CEFR word list

Why a median

Word difficulty is less settled than it looks. Before Oxford 5000 became one of our sources, we compared our earlier list with it on the 4,350 words they shared: the two agreed exactly on 39% and to within one step on 84%. Two careful authorities disagree on the majority of shared words, so treating any single list as ground truth overstates what is known.

Taking the median of several sources does not remove that uncertainty, but it stops any one source from setting the scale on its own, and it makes the uncertainty measurable: we keep the spread between the highest and lowest opinion for every word.

Some entries keep the form in which a source printed them, including fragments of phrases (“(sth)”, slashes); we are cleaning these.

What we do with disagreement

851 words carry a spread of two steps or more. Most are not errors but polysemy: minute is A1 as a unit of time and C1 meaning “negligible”; net is elementary as an object and advanced as net profit. Our list is keyed on spelling and cannot tell those apart, so we exclude disputed words from our own level measurements. Measured effect: including them dropped our text-level accuracy from 66% to 58% (our previous dictionary scored 65% on the same texts).

Sources, licences and what is not here

CEFR-J vocabulary profile, copyright Tono Laboratory, Tokyo University of Foreign Studies. Free for research and commercial use provided the dataset is cited. Octanove Vocabulary Profile C1/C2, CC BY-SA 4.0. Academic Word List (Coxhead, 2000) is used elsewhere in our measurements, not in this list. Ministry of Education lexical bands come from the Israeli English curriculum.

Licence of this list. This word list is an adaptation of the Octanove Vocabulary Profile C1/C2, which is licensed CC BY-SA 4.0. Under section 3(b) of that licence an adaptation must be released under the same terms, so this list is published under the Creative Commons Attribution-ShareAlike 4.0 International licence: copy it, redistribute it and build on it, including commercially, provided you credit the sources and license what you build under the same licence. Full licence text: creativecommons.org/licenses/by-sa/4.0/legalcode.en.

Words marked (Ox) are words where Oxford 5000 was one of the sources consulted. The published level for those words is still our median, not Oxford's value.

604 words are deliberately absent. For those, Oxford 5000 was the only source, so their level would be Oxford's own assignment rather than our derived one. Oxford 5000 is copyright Oxford University Press and we have not established a right to redistribute it, so we do not. We would rather publish a smaller list we can stand behind than a larger one we cannot.

This list is a working artefact of the YAKKI text-levelling work, published so others can check it. If you hold rights in any material here and believe something should not be published, write to [email protected] and we will remove it.